Chris wrote:
On 20/01/2019 18:07, Filip454 wrote:
On 2019-01-19 10:43, Chris wrote:
Mr. Man-wai Chang wrote:

Can a class-action suit be filed to US Court to force Micro$oft to
extend its support for Windows 7?

Based on what? It's practically a 10 year old os that had been
superseded
twice (three times, if you include 8.1) and Microsoft has very
clearly and
in plenty of time announced its EOL.

So not a chance.

As others have said though there's nothing to stop you from
continuing to
use it.

I do NOT think Windows 10 is ready for the prime time, it's update is
absolutely as horrifying as a data doomsday.

It's a horrible os, but it does work as well as any windows release.




I do not agree at all.

Especially on slower laptops, Windows 10 runs absolutely TERRIBLE -
cheap Pentium laptops for example like HP G4 250. It should not be
even preinstalled on those machines.

Windows has never worked well on under-powered (for it's generation)
hardware.

My "works as well as as any windows" is a very back-handed comment. It
means that when it works it works fine, but sometimes it can be crashy
as hell for people for no real reason.


I've benched hardware under the various OSes, and
there really isn't much difference at the CPU cycle
level. Since Windows 10 "reserves" cycles to remain
responsive, you have to "oversubscribe on threads" to
get 100% of the CPU capability. For example, on a 4 core
CPU, you might use 8 threads in 7ZIP ultra, to drive the
CPU to 100% instead of 85-90%.

By doing some careful benchmarking, on "relatively simple"
architectures, there's no difference. Windows 7 is ahead
by maybe 1%, but it could easily be "poor technique" on
my part or measurement error. I repeated some of my work
three times (cache warmup or whatever), just to make sure
I wasn't doing it wrong. You either cache warmup, or you
reboot before every test sequence to re-establish initial state.

When the CPU arch is "squirrel strange", like on a ThreadRipper
or Epyc, more holes could show through. Embarrassing holes,
that make it apparent the CPU isn't tuned by any OS loaded
on it. When facilities in a CPU are not uniform, this is
what happens. (Even though Microsoft has moaned and groaned
for several of their OSes, that "things were better".)

AMDs next version, the one that uses chiplets, hopes to get
around this issue by making things more symmetric, more of
the time.

On an Intel processor with the dual rotating rings, you
lose around one core of performance due to the bus. A six
core processor gives five cores of thruput. Intel attempted
to fix that using mesh busses (a bus array), but I've not
examined any results on the web to see whether that improved
things or not. And you got "more of your moneys worth".

The interconnect issue was present in previous generations.
A Q6600 which consists of two dual-core dies in the same
CPU package, gives around 3.5 cores of performance out
of a max of 4 cores. This is due to snoop traffic on the
FSB - both cores share the FSB with the Northbridge
connection, and "chit-chat" between processors robs
the design of a bit of performance. Once the memory
controller and PCIe video were brought inside the
CPU, and the CPU consisted of one core, the efficiency
came back.

Usually a 4C8T processor is small enough now, to be
"devoid of squirrels". Anything larger, do the research.

And integrated video is another issue. There is at least
one SOC that had a strange GPU in it, and only one driver
was made for it. Which in a rolling release like Windows 10,
is a liability (since the WDDM version number could be bumped).

Paul